Breast cancer is a complex and heterogeneous disease with a variety of histopathological and molecular subforms with diverse clinical outcomes and relationships with established risk factors 1 2 . One important subclassification of clinical breast tumors is based on the presence or absence of estrogen (ER) and progesterone (PR) receptors, and the routine identification of these receptors currently guides targeted therapies and provides important prognostic information 3 . The expression of the hormone receptors also broadly overlaps with more detailed molecular subclassifications of breast tumors as determined by microarray-based gene expression profiling 4 5 .

Epidemiological data indicate that associations between excess body mass index (BMI) and the risk of breast cancer may differ by the ER and PR status in breast tumors 6 , and that the positive association of excess adiposity with breast cancer risk after menopause may be driven predominantly by the association with receptor-positive (ER+ or ER+PR+) disease 6 . For receptor-negative breast tumors, the association of risk with excess weight is less well characterized, and studies have shown variable and somewhat conflicting results 6 7 8 9 . In part, the diversity of previous findings could be related to heterogeneity in the association of excess weight with breast cancers diagnosed at a predominantly premenopausal or more advanced postmenopausal ages.

Another important factor to be accounted for when assessing the relationships of excess adiposity with postmenopausal breast cancer risk is use of hormone replacement therapy (HRT). Various studies have shown interactions of BMI with postmenopausal HRT use as risk factors for breast cancer 6 10 11 12 13 14 . However these interactions have not been described for hormone receptor-defined breast cancer subtypes.

The present analysis extends on a previous report (2004) from the European Prospective Investigation into Cancer and Nutrition (EPIC) in which the relationships of anthropometry with risk of total breast cancer were investigated 14 . Using more recent follow-up data from EPIC, which now includes a total of 7,174 incident cases of breast cancer with information on ER status and 5,906 with additional PR status, we examined the relationships of anthropometric indices of adiposity (BMI, waist and hip circumferences) with risks of breast cancer subtypes defined by hormone receptor status, across five-year age bands spanning from premenopausal into postmenopausal years, and accounting for past or current use of menopausal HRT.

EPIC is a multi-center prospective cohort study designed to investigate the relationships between diet, nutrition and metabolic factors and cancer, consisting of about 370,000 women and 150,000 men aged mostly between 25 and 70 years 15 16 . All participants were enrolled between the years 1992 and 2000 from 23 regional and national research centers located in 10 western European countries: Denmark, France, Germany, Greece, Italy, The Netherlands, Norway, Spain, and the United Kingdom.

Extensive details about the standardized procedures for recruitment, measuring baseline anthropometry (height, weight, waist and hip circumferences), questionnaires on current habitual diet, reproductive and menstrual history, exogenous hormone use (oral contraceptive (OC) and HRT use), medical history, lifetime smoking and alcohol consumption history, occupation, level of education and physical activity and biological sample collection at study centers are given elsewhere 15 16 . All subjects gave written informed consent to use their questionnaire data and the Internal Review Boards (IRB) of International Agency for Research on Cancer (IARC) and all EPIC recruitment centers approved the analyses based on EPIC participants.

A total cohort of 314,760 women was followed for a sum of 3,399,178 person years. Of these, 144,223 (45.8%) were postmenopausal at the time of recruitment, of which 93.6% had a natural menopause. Baseline characteristics of all women and postmenopausal women within this cohort are presented in Table 1. The median age at recruitment from the EPIC centers was 50.9 years for all women, 58.1 for women who were postmenopausal, 42.6 for women who were premenopausal and 50.1 for women whose menopausal status was either perimenopausal or undetermined at baseline. Of the postmenopausal women who provided information on HRT use, 42,719 (29.6%) were current users at recruitment, 18,278 (12.7%) were past users and 78,919 (54.7%) were never users. Postmenopausal women of higher degrees of BMI were more likely to be never users of HRT and had a shorter duration of past HRT use (Additional file 1 Table S1).

For the great majority of postmenopausal women who reported HRT current use and had information of HRT regimen (n = 35,856 (76.0%)), combinations of estrogen plus progestin were used, while almost one out of four HRT users (24.0%) used formulae based on estrogens alone. Almost half of all breast cancer cases (46.2%) were diagnosed at the age of 60 years or older. The median time until diagnosis was 6.1 years for all incident cases, 6.3 years for ER-PR- tumors, 6.8 years for ER+PR+ tumors, and 5.0 years for tumors with unknown ER or PR receptor status. Incident cases with ER-PR- breast tumors were rather evenly distributed across the age bands, whereas the majority of cases with ER+PR+ breast tumors were diagnosed after the age of 55 (Table 2). A basic calculation of incidence rates across the five-year age bands showed that ER-PR- breast cancer rates were stable across all age bands above age 50 (for ≤49 years: 16.9, 50 to 54 years: 34.6, 55 to 59 years: 36.9, 60 to 64 years: 38.4, ≥65 years: 29.1 per 100,000 person years), whereas ER+PR+ disease, rates increased steadily with increasing age albeit at a slower rate after 50 years (for ≤49 years: 44.5, 50 to 54 years: 111.2, 55 to 59 years: 125.4, 60 to 64 years: 142.6, ≥65 years: 127.9 per 100,000 person years). 28

Across the age bands, Cox regression models showed no distinct association for BMI with risk of ER-PR- breast cancer (Figure 1 and Table 3). It was only among women older than 65 years that the relative risk for ER-PR- tumors (per five-unit increase in BMI HR = 0.95 (95%CI: 0.79 to 1.13)) was statistically different from ER+PR+ tumors (per five-unit increase in BMI HR = 1.25 (95%CI: 1.16 to 1.34); P < 0.001; P_het = 0.004). For ER+PR+ tumors, the negative risk association for BMI among women aged less than 49 years (per five-unit increase in BMI HR = 0.79 (95%CI: 0.68 to 0.91); P = 0.002) progressively turned into a significant positive risk association among women aged 65 years and older.

Restricting the analysis to women who never used HRT, a five-unit increase in BMI showed a weak tendency towards a risk of ER-PR- disease; although this trend was not statistically significant across any age band. There was an indication of heterogeneity between the risks estimates of ER-PR- (per five-unit increase in BMI HR = 1.06 (0.85 to 1.32)) and ER+PR+ tumors (per five-unit increase in BMI HR = 1.34 (1.22 to 1.47)) among the HRT never users older than 65 years (P _het = 0.05). Similar patterns of association were observed for BMI with risk of ER-negative versus ER-positive breast cancer subtypes, ignoring PR status (Additional file 2, Figure S1), and as well as PR-negative versus PR-positive breast cancer subtypes, ignoring ER status (Additional file 3, Figure S2).

Increased waist or hip circumferences showed relative risk patterns across the age bands that were very similar to those for BMI, for both ER+PR+ and ER-PR- breast cancer subtypes, among all women as well as among current and past users of HRT (Additional files 4 and 5, Figures S3 and S4). After adjustment for BMI, however, neither waist nor hip circumferences showed any relationship with risks of either breast cancer subtype (data not shown). Pearson's correlations showed that increases in BMI were highly correlated with increases in both waist and hip circumferences (Pearson's correlations coefficient r = 0.83 and 0.85 respectively), therefore this could account for the lack of associations observed of waist and hip circumferences after adjustment for BMI.

With regard to postmenopausal HRT use, Cox regression models showed that in comparison to HRT never users, postmenopausal women who were current HRT users at baseline had significantly increased risks of both ER-PR- breast cancer (HR = 1.30 (95%CI: 1.05 to 1.62); P _value = 0.02) and ER+PR+ breast cancer (HR = 1.74 (95%CI: 1.56 to 1.95); P _value < 0.001), although the relative risk was significantly weaker for ER-PR- than for ER+PR+ tumors (P _het = 0.035).

For combined risk categories determined by the joint classification of HRT use (never, past, current) and BMI tertiles, for ER-PR- breast cancer, using never users of HRT in the lowest BMI tertile as a reference category, the highest relative risk for current HRT use was seen for women in the first BMI tertile, while relative risks were somewhat weaker for women in the middle and upper tertiles of BMI (Figure 2). Within separate categories of HRT use, we observed a positive risk association within the third tertile of BMI with ER-PR- breast cancer among never users of HRT (third versus first tertile HR = 1.47 (95%CI: 1.01 to 2.15); P _trend = 0.06) (Table 4) contrasted by an indication of a negative risk association among current and no association with past users of HRT. Switching perspective and stratifying the analysis by separate strata of BMI, the relative risk for current HRT use was strongest for women within the first BMI tertile and absent for women in the higher tertiles. This interaction of the effects of HRT use (current versus never users) across BMI tertile scores was statistically significant (P _interaction = 0.037) (Table 5) and was not restricted to HRT type (Additional files 6, 7, 8, 9, 10 and 11, Tables S2 to S5 and Figures S5 and S6).

For ER+PR+ breast cancer, compared to never users of HRT in the lowest BMI tertile as a reference category, we observed an approximate 2.5-fold increase in risk for ER+PR+ breast cancer for baseline HRT users in each of the three BMI tertiles when we used never users in the lowest BMI tertile as the reference category (Figure 2). However, performing analyses within independent strata for baseline HRT use (never, past, current), our data showed more than a 1.8-fold increase in risk for the highest BMI tertile compared to the lowest tertile in the strata of never and past users of HRT, but no such association at all among current HRT users (Table 4). Alternatively, examining relative risks for HRT use within strata of each of the BMI tertiles separately, the relative risk for ER+PR+ breast cancer associated with current use of HRT was strongest for women within the lowest BMI tertile, and progressively weakened across BMI tertiles 2 and 3, respectively (P _interaction < 0.001)(Table 5).

Overall, for ER-PR- tumors, our study showed no distinct associations with BMI in any of the five-year age bands. However, among postmenopausal women who never used HRT, our data showed a significant BMI-related increase in risk of receptor-negative breast cancers. As expected, the inverse relationship of excess body weight with risk of ER+PR+ breast cancer among women younger than 49 years progressively turned into a direct relationship among women of more advanced age. Furthermore, we observed increases in the risks of both hormone receptor-negative and -positive tumors among women who used HRT at baseline, although the increase in risk was weaker for ER-PR- than for ER+PR+ tumors. The HRT-related increases in risks of both ER-PR- and ER+PR+ breast cancer were stronger among the leaner women, and this increase was independent of HRT type. Finally, with respect to measures of body fat distribution (waist and hip circumferences) no relationships with either receptor-negative or -positive breast cancer were observed after adjustment for BMI.

Cancer registry data show that overall breast cancer incidence increases with age. However, after the age of 50, rates of hormone receptor-negative disease no longer increase with advancing age, whereas the incidence rates of ER-positive tumors continue, albeit at a reduced pace compared to earlier ages 29 30 31 32 33 . This age-related pattern of incidence rates of ER-PR- and ER+PR+ tumors was also apparent in our data. The change in the age-related rise in incidence rates after age 50 suggests a possible relationship of both receptor-negative and -positive breast cancer with the menopause-related cessation of ovarian estrogen and/or progesterone synthesis 29 . While a wide array of epidemiologic, clinical and experimental evidence has clearly established a late-stage growth-promoting effect of estrogens, especially on estrogen-sensitive tumors 34 , there is substantial evidence that estrogens may also play an important role in earlier evolutionary stages of the development of both ER-negative and -positive tumor types 29 35 36 . Indeed, evidence suggests that a large proportion of ER-negative tumors may arise from estrogen-responsive precursor tumors or cells and that estrogen sensitivity may be lost at later stages of tumor development 35 . In addition, mammary stem cells have been shown to be responsive to sex steroid hormones despite not having a clear expression of an ER or PR 37 .

For ER-PR- tumors, our findings across the age-bands are consistent with those from the meta-analysis by Suzuki et al. (2009) 6 of nine cohort and twenty-two case control studies, which also showed no association of BMI with risk of ER-PR- tumors in either premenopausal (1481 cases) or postmenopausal women (1522 cases). However, when our analysis was restricted to postmenopausal women who never used HRT, BMI was associated with ER-PR- tumors. A recent analysis of the Nurses' Health Study II cohort showed a more than two-fold increase in risk of ER-negative premenopausal breast cancer (n = 131) for women in higher categories of BMI, waist and hip circumferences 38 . In addition, an analysis within the Women's Health Initiative cohort 39 showed a moderate increase in risk of triple-negative (ER-/PR-/HER2-) breast tumors among postmenopausal women who had more elevated BMI (HR = 1.35 (95%CI = 0.92 to 1.99) for the top versus bottom quartiles of BMI).

For receptor-positive tumors, our findings on the relationships of risk with BMI are well in line with those from previous epidemiologic studies. The meta-analysis by Suzuki et al. showed a 33% increase in risk of ER+PR+ tumors per five-unit increase in BMI among postmenopausal women (5,469 cases; risk estimate (RE) = 1.33 (95%CI = 1.20 to 1.48)), and a 10% reduction in risk of ER+PR+ disease among premenopausal women (2,643 cases; RE = 0.90 (95%CI = 0.82 to 0.99)) 6 . Furthermore, similar to our observations, this meta-analysis showed a somewhat stronger association of BMI with risk of hormone ER+PR+ tumors for women who did not use HRT 6 . Finally, still in line with our findings, one recent prospective study also showed an inverse relationship of BMI with breast cancer risk in premenopausal women that was restricted to ER-positive tumors 40 , and another recent study showed that the direct risk association of BMI with ER-positive cancers was stronger among postmenopausal women who did not use HRT 39 .

The direct risk association of general obesity with both ER-PR- and ER+PR+ breast cancer in postmenopausal women not using HRT, may be explained by adiposity-related increases in circulating estrogen levels, due to conversion of androgens into estrogens within adipose tissue 41 . The restriction of BMI-related risks to HRT nonusers is plausible, because HRT use is a dominant source of circulating estrogens and progestins in postmenopausal women 42 that augments blood concentrations to levels equaling those in premenopausal women, compared to which adiposity-related effects on endogenous synthesis are small.

The mechanisms that underlie the inverse association of BMI with risk of ER+PR+ tumors among women of reproductive age are still largely unclear 43 . Although it has been speculated that obesity-related reductions in circulating estrogen levels and/or reductions in circulating progesterone (due to chronic anovulation) could play a role 44 , epidemiological studies of breast cancer risk in relation to premenopausal blood levels of progesterone and estrogens provide no direct support for this 44 45 .

An important observation in our study is that current users of HRT, compared to women who did not use HRT, had higher relative risks of both receptor-negative and -positive, but with a stronger increase in risk for the ER+PR+ subtype. Some previous prospective studies have documented similar increases in risks of both ER-negative and -positive breast tumors 46 47 , especially among users of combined estrogen-plus-progestin regimens 47 48 49 , whereas other studies showed such increases only for ER-positive disease 9 50 .

Another significant observation in our study was that for both breast cancer subtypes the increased risk association in current HRT users depended upon a women's BMI. For receptor-negative and -positive breast cancers, HRT-related increases in risk were stronger among the leaner women and was not restricted to a particular HRT type. Switching perspectives, this statistical interaction also manifested itself in a lack of BMI-related breast cancer risks for both receptor-negative and -positive tumors among current HRT users. A previous analysis of the EPIC data 14 and several other studies 6 10 11 12 13 50 51 have shown an inverse association of BMI with the risk of breast cancer overall among HRT users and, from our present analysis, it appears that this overall inverse relationship could be attributable to a relative attenuation of the HRT-related increase in risk in receptor-negative and -positive tumors among more obese women. The endocrine or other mechanisms that may underlie this attenuation of the HRT-related risk remains elusive.

The stronger association of HRT use in lean women in comparison to more overweight women could be potentially related to the different patterns of HRT use. As expected, we observed, a higher proportion of current use in lean women. Within past users of HRT, women of a higher BMI tended to have a shorter cumulative duration of HRT use. Although data was restricted to information collected at baseline, risk estimates in past and current users were not affected by adjustment for duration of use (data not shown).

It is important to acknowledge the limitations of our study. The determination of ER and PR status in breast tumors is a standard part of breast cancer diagnosis used to predict endocrine therapy response 3 . Nevertheless, while a number of studies have shown that the classification of the ER and PR in tumors is relatively robust 52 53 , the accuracy of classifying an ER- or PR-negative tumor remains controversial 54 55 . Within the EPIC cohort, as in most other studies, the classification of breast cancer as receptor-positive and -negative (ER, PR or joint ERPR) is based upon a variety of laboratory assay types and quantification methods, reflecting pathology practices across Europe and time. Recently, recommendations were made to lower the threshold of the percentage of stained cells from ≥10% to ≥1% to indicate receptor-positive tumors 56 57 , and some of our breast cancer cases classified as ER-negative or PR-negative might have been actually receptor-positive tumors according to these newest criteria. A further limitation of our study is the lack of prospective follow-up information of both BMI and HRT use after recruitment.

Major strengths of the present study are its prospective design and large number of incident cases with receptor information. This large number of cases allowed in-depth analyses of BMI-related relative risk patterns across age bands, describing risk associations and their progressive changes from premenopausal to late postmenopausal age. Furthermore, the large case numbers allowed the examination of interaction effects between excess body weight and use of postmenopausal HRT and by HRT type.

In summary, we found for receptor-negative tumors, increased BMI among postmenopausal women was associated with risk of receptor-negative tumors, however, only in never users of HRT. In previous studies, excess body weight was associated with an inverse risk only of receptor-positive breast cancer in women aged less than 49, which progressively turned into an increased risk of receptor-positive disease as women aged. Furthermore, women who reported current HRT use at baseline were at an increased risk of both receptor-negative and -positive breast cancer, although this increase was stronger for receptor-positive tumors. In both receptor-negative and -positive breast cancers our data show a stronger HRT-related increase in risk among leaner than among more obese women. This latter interaction also meant that BMI-related increases in risks of receptor-negative and -positive breast cancers were observed among nonusers of HRT only, whereas a possible BMI-related relative decrease (receptor-negative disease) or no increase in risk (receptor-positive disease) may be seen among current users of HRT.

BMI: body mass index; CI: confidence interval; EPIC: European Prospective Investigation into Cancer and Nutrition; ER: estrogen receptor; HR: hazard ratio; HRT: hormone replacement therapy; OC: oral contraceptives; PR: progesterone receptor; RE: risk estimate.

The authors declare that they have no competing interests.

RR, AL, LD and RK contributed to the conception of the current analysis and all authors were involved in the design and acquisition of data from the EPIC cohort. RR, AL, LD and RK contributed to the analysis and all authors contributed to the interpretation of the data. RR, AL and RK drafted the manuscript and all authors revised the final draft critically for important critical content. All authors have given final approval of the version to be published.